Overseas market information and firms' export decisions.

Inui, Tomohiko ; Ito, Keiko ; Miyakawa, Daisuke 等

I. INTRODUCTION

The relationship between globalization and firms' performance has been the subject of numerous studies, which have tended to find that there is a positive relationship between the two. Yet, researchers' understanding of the dynamic behavior of firms in a globalized economy is still far from sufficient to propose specific policies that help firms to grow in such an environment. For instance, micro-data analyses on various countries confirm that the international performance of a country tends to hinge on a handful of high-performing firms (Mayer and Ottaviano 2008), suggesting that increasing the number of firms involved in international activities is important for the successful internationalization of a country. However, both theoretical and empirical research to date have not produced an adequate answer to the question of how to increase the number of firms involved in international activities. For example, although there is wide empirical support for the theoretical prediction that firms with higher productivity are more likely to become exporters, a growing number of studies is producing results suggesting that productivity advantages alone do not sufficiently explain the self-selection of firms into exporting. Such studies (see, e.g., Bernard et al. 2003; Mayer and Ottaviano 2008; Todo 2011) point out that although such productivity advantages certainly do appear to exist, their impact is economically negligible. This implies that our knowledge about the determinants of the export decision remains very limited, and no conclusive answer has yet been found as to what factors are important for firms to become an exporter and grow through exporting.

The extant literature has focused on a number of conditions or factors that may affect firms' export decision. One important research strand in this context concentrates on export spillovers. The idea is that information exchange with other exporting firms reduces the individual fixed costs associated with exporting, and that such information exchange therefore increases the probability that a firm will export (see, e.g., Krautheim 2012, for a theoretical analysis). The hypothesis suggests that having access to information on foreign markets substantially reduces uncertainty and encourages firms to engage in export activities. Empirical work by Koenig, Mayneris, and Poncet (2010) confirms this hypothesis by finding that the presence of other exporters has a positive effect on the export decision of other firms. Although Koenig, Mayneris, and Poncet (2010) find evidence of positive export spillovers, the evidence produced by other empirical studies on such export spillovers is at best weak (e.g., Aitken, Hanson, and Harrison 1997; Barrios, Gorg, and Strobl 2003; Bernard and Jensen 2004), which means that the search for possible channels of information spillovers continues.

Against this background, this study focuses on information provided by lender banks as one potential channel of information spillovers. Most existing empirical studies examining information spillovers from other exporting firms assume that firms in the same region and/or industry are likely to exchange information with each other; however, such studies do not explicitly discuss the channel through which such information exchange takes place. The hypothesis we examine here is that lender banks work as a conduit for such information. While there are likely a variety of information channels, the reasons that we focus on the exchange of information through lender banks in particular are as follows. First, as the monitoring of borrower firms is important for banks, one would expect banks generally to accumulate information on borrower firms and related parties. Thus, if we assume that a particular bank is very knowledgeable about overseas business opportunities either through its own banking activities or transactions with client firms with experience in exporting, potential exporter firms would find it helpful to consult with such a bank. Second, it has been widely pointed out that in the case of Japan, lender banks provide not only financial support but also business consulting services utilizing the extensive knowledge collected through their lending transaction relationships and from various information sources. This is particularly the case for firms' "main bank," which is usually the bank from which a firm has borrowed the most and with which it has a long-term relationship, and which typically plays an important role as a provider of both funds and information. As argued by Patrick (1994), such long-term relationships between main banks and borrower firms enable banks to gain access to "soft" information on borrower firms, which helps to raise the efficiency of loan screening and borrower monitoring. (1) Although other various benefits as associated with parent firms, corporate groups (conglomerates), and related firms in same corporate groups and/or upstream and downstream firms with transaction relationships are of course important (e.g., Aw et al. 2000; Rodrik 1995), we do not explicitly consider these channels mainly due to data limitations. Accordingly, we should also note that many Japanese corporate groups were traditionally bank centered, wherein their main banks were able to accumulate various information on firms belonging to the corporate groups. Even non-bank centered corporate groups usually have their main banks, and many firms belonging to the same corporate groups use the main same bank. Although inter-firm and firm-bank relationships in such corporate groups have been weakened and transactions beyond corporate groups have been increased since the late 1990s, most corporate groups' main banks have continued to serve as main banks for their former group firms. This historical background allows us to focus on the main bank-firm relationship in our analysis. Nevertheless, inter-firm transaction relationships within and across corporate groups are important information channels and they deserve scrutiny for future research.

The aim of this paper is to explore the role of banks as information providers by explicitly quantifying banks' ability to provide information on export markets using a unique panel data set for Japan in which firms are matched to lender banks. We therefore conjecture that banks play a crucial role in substantially reducing the fixed entry costs incurred by client firms when starting to export. Specifically, we hypothesize that the provision of information by lender banks helps firms to start exporting based on the same mechanisms that information exchange with other exporting firms helps potential export starters. To examine this hypothesis, we focus on firms' main bank which, in line with previous studies, we define as the top lender bank of a firm and investigate the importance of information flows from the main bank to client firms as a source of spillovers. (2)

As far as we know, this is the first empirical study to examine the impact of information spillovers through main banks on both firms' decision to start exporting (the extensive margin) and on the volume exported by each firm {the intensive margin). This contrasts with the focus of most of the previous literature on the role of banks, such as the studies by Amiti and Weinstein (2011) and Paravisini et al. (2015), which examine the relationship between banks' financial health and their clients' export behavior. This study thus contributes to the literature by adding a new element to the determinants of firms' export decision, namely the role of bank information. Our results show that information on overseas markets provided by a main bank substantially reduces the fixed costs of starting exporting for a firm and thereby increases the probability that the firm will start exporting. However, the effect of such information on the volume of exports is not very clear.

This paper is organized as follows. Section II presents the empirical strategy. Section III describes the data set used in this paper and provides some descriptive statistics on our sample firms. Section IV presents our estimation results. Finally, Section V discusses the policy implications and concludes.

II. EMPIRICAL STRATEGY

This section explains the empirical strategy we employ to investigate the determinants of the export decision and of the export volume. We are particularly interested in the impact of information provided by main banks on the probability that a firm starts exporting (i.e., the extensive margin) and on the export volume (i.e., the intensive margin). Following previous empirical studies on the determinants of the extensive and intensive margin (e.g., Koenig et al. 2010; Minetti and Zhu 2011), we assume that firm i starts exporting if its profits are larger when exporting than when not exporting. Let [[pi].sup.*.sub.ijt] represent the difference between the profits of firm i when it starts exporting to destination j at time t and its profits when it does not start exporting to destination j at time t. The difference is determined by firm characteristics (e.g., size, productivity, and the skill level of workers), the firm's financial conditions (e.g., the leverage ratio, liquidity ratio, and short-term loan ratio), and the amount of information on the export market available to the firm. The availability of information on the export market is assumed to substantially lower the uncertainty of profits from exporting and hence to lower either the variable or the fixed cost of exporting. While export spillovers are also taken into account, we are particularly interested in information provided through the main bank of the firm. Therefore, we parameterize [[pi].sup.*.sub.ijt] as

[[pi].sup.*.sub.ijt] = [[alpha].sub.1] + [Z.sub.it] [[beta].sub.1] + [I.sub.ijt] [[gamma].sub.1] + [[epsilon].sub.ijt],

where [Z.sub.it] is a vector of controls for firm characteristics and the firm's financial conditions that may affect firm i's differential profits [[pi].sup.*.sub.ijt]; [I.sub.ijt] is a vector of variables representing information available to the firm; and [[epsilon].sub.ijt] captures unobserved firm characteristics and other unknown factors that may also affect differential profits.

We assume that firm i starts exporting if the differential profits [[pi].sup.*.sub.ijt] > 0. Under the assumption that [[epsilon].sub.ijt] is a normally distributed random error with zero mean and unit variance, the probability that firm i starts exporting can be written as (1)

[Prob.sub.ijt] = Prob ([[alpha].sub.1] + [Z.sub.it] [[beta].sub.1] + [I.sub.ijt], [[gamma].sub.1] + [[epsilon].sub.ijt] > 0).

In the first instance, we estimate Equation (1) with a random effect panel probit approach. In order to take any potential endogeneity into account, we lag all right-hand-side variables by 1 year. The dependent variable [Prob.sub.ijt] denotes the change in export status at the firm- or firm-destination level and takes a value of 1 if a firm exports for the first time (overall) or the first time to destination j at time t. We define a firm as an export starter if the firm did not export over the last 3 years from t - 3 to t - 1 and exports at time t. [Prob.sub.ijt] takes a value of 0 if a firm did not export to destination j for the last 3 years prior to year t and does not export in year t. Firms that always export to destination j are not included in our analysis. Regarding control variables for firm characteristics and the firm's financial conditions ([Z.sub.it]), we include firm size (the log of the number of employees of firm i), the total factor productivity (TFP) level of the firm, and the average wage rate of the firm as a proxy for the skill level of workers. Based on the results of both theoretical and empirical studies, we expect these variables to be positively correlated with firms' export decision. Furthermore, to take the impact of liquidity constraints on firms' export behavior into account, we include variables representing firms' financial situation, such as their leverage ratio, their liquidity ratio, and the share of short-term loan in their total loans outstanding. The reason for including these variables is that, as highlighted by, for example, Minetti and Zhu (2011), Feenstra et al. (2014), and Manova et al. (forthcoming), financial constraints are likely to prevent firms from exporting because firms need sufficient liquidity in order to meet the entry costs associated with starting exporting. Therefore, we expect that firms with more liquidity are more likely to start exporting.

Regarding information available to the firm ([I.sub.ijt]), we include variables representing the amount of information on export markets accumulated by a main bank and by a firm itself. The explanatory variable of main interest is the amount of information on export markets potentially available to the firm through its main bank, which is a proxy for the amount of information firm i's main bank has accumulated on destination j. Specifically, we measure this variable as the ratio of the number of the main bank's client firms that are exporting to destination j to the total number of the main bank's client firms, that is, the intensity of each main bank's dealings with exporting firms. We conjecture that banks with more extensive dealings with exporter firms more effectively accumulate information related to overseas markets. This could be the case when, for example, banks allocate limited lending/managerial capacity to each lending activity. Under such circumstance, a higher intensity of dealing with exporting firms represents the extent to which the bank focuses (i.e., allocates internal resources with higher priority) on the lending activities accompanied by the provision of overseas market information. In order to take into account the information accumulated by firms themselves through their own international activities, we also include variables representing their overseas activities, such as the share of overseas employees in a firm's total number of employees and the share of overseas investment in a firm's total investment. Industry dummies (for 15 manufacturing industries) and time dummies are also included in order to control for industry-specific and time-specific fixed effects. As highlighted in previous studies, there may be some spillovers from nearby exporters. In order to examine whether this is the case, we included dummies for the region in which firms' headquarters are located in order to control for export spillovers and other region-specific factors. However, we found that the region dummies were not significant and including them did not increase the explanatory power of our results, so that we decided to omit them here. A possible reason is that the headquarters of most firms in our sample are concentrated in a small number of prefectures (Tokyo, Osaka, and Hyogo prefectures).

While Equation (1) focuses on the extensive margin, that is, whether firms start exporting, we also examine the role of information spillovers through the main bank on the intensive margin, that is, the export volume after firms start exporting. To do so, we adapt Equation (1) above as follows:

(2) [EXP.sub.ijt] = [[alpha].sub.2] + [Z.sub.it] [[beta].sub.2] + [I.sub.ijt] [[gamma].sub.2] + [[epsilon].sub.ijt],

where [EXP.sub.ijt] is the log of firm i's exports to destination j at time t. We also use the first difference of the log of exports (i.e., the growth rate of exports) as a dependent variable for an alternative specification. The variables on the right-hand side are the same as those in Equation (1) and we again lag all variables by 1 year. As above, the variable we are most interested in is the amount of information on export markets potentially available to the firm through its main bank.

That the provision of information by the main bank may affect not only the extensive margin but also the intensive margin is suggested by the theoretical analysis by Rauch and Watson (2003), who examine the relationship between the search costs for establishing new partnerships and export volumes. They suggest that the higher the costs of searching for a new supplier, the smaller tend to be the orders a buyer places with a supplier. In addition, buyers tend to place larger orders with suppliers once they know that the latter is able to fulfill larger orders. Based on this idea, if banks help in matching businesses in overseas markets and provide information to both the buyer and the supplier on their respective counterpart, this should substantially reduce uncertainty and possibly result in higher transaction volumes. We test this hypothesis by examining whether information spillovers through the main bank have a positive effect on the intensive margin or not. Note that in the estimation of Equation (2), non-exporters are excluded from the sample used for analysis. In cases such as here, where there is a risk of a selection bias, a typical solution employed often is to use a Heckman selection model. However, we do not employ the Heckman model and estimate Equations (1) and (2) separately, as it is difficult to find a variable that strongly affects the selection process (Equation (1)) but not the outcome (Equation (2)). Therefore, we estimate Equation (2) separately from Equation (1), employing the fixed-effect panel estimation method.

III. DATA AND DESCRIPTIVE STATISTICS

A. Data Description

The data used in this study are the firm-level panel data from the Basic Survey on Business Structure and Activities (BSBSA) collected annually by Japan's Ministry of Economy, Trade and Industry (METI) for the period 1997-2008. The survey is compulsory and covers all firms with at least 50 employees or 30 million yen of paid-in capital in the Japanese manufacturing, mining, wholesale and retail sectors, and several other service sectors. The survey contains detailed information on firm-level business activities such as the three-digit industry in which the firm operates, its number of employees, sales, purchases, exports, and imports (including a breakdown of the destination of sales and exports and the origin of purchases and imports). (3) It also contains R&D expenditures and patents owned, the number of domestic and overseas subsidiaries, and various other financial data such as costs, profits, investment, debt, and assets.

The key aim of our analysis, as mentioned above, is to investigate the importance of information on destination markets and advice provided by main banks to their client firms. To do so, we combine the firm-level data with information on firms' main bank and examine the relationships between firm characteristics, main banks' ability to provide advice, and firms' export status. We augment the firm-level panel data taken from the BSBSA with information on firm characteristics stored in the Nikkei Economic Electronic Database System (NEEDS) Financial Quest database. We then merge the data set with information on the main bank for each firm using the loan relation information stored also in the NEEDS Financial Quest database. This database also includes various types of information on main banks.

While the BSBSA includes a large number of unlisted firms, we restrict our sample to listed firms in this paper. This reflects our intention to control for firms' characteristics as comprehensively as possible. As the NEEDS Financial Quest database contains all the information on firms' financial statement, we can control a variety of firm characteristics. This is especially important for our analysis, given the potential risk of endogeneity problem due to omitted variable associated with firms. Even though we limit our sample to listed companies so that we can match firms to their main bank, our data set nevertheless includes a considerable number of small firms, which are listed on the stock exchange markets for start-up companies, and some of them are first-time exporters. Moreover, once firms have started exporting, many of them expand the range of destinations to which they export, so that when we examine the determinants of whether firms start exporting to a new destination, we can include more observations in our analysis.

Our unbalanced panel data contain 300-400 listed firms per year, 5% of which are identified as export starters. We were able to match the BSBSA data with the other two databases for 9,300 observations in the manufacturing sector. However, the sample size for our analysis is at most 3,000 observations. The reasons are as follows. First, we exclude firms that have positive exports throughout our observation period ("always" exporters), as our focus is on the decision to start exporting. Second, firms for which data on bank loan transactions are not available are excluded from our data set. Third, as we employ a 3-year window for identifying first-time exporters, firms that frequently changed their export status are excluded from our data set. Specifically, in our analysis, export starters are defined as firms that started exporting in year t but did not export in years t - 3 to t - 1. Although the number of pure first-time exporters is limited, there are several exporters that expanded or reduced the number of destinations to which they exported during our observation period.

B. Variables

Let us now describe the variables for our estimation in detail. The basic statistics of all variables are provided in Table 1. Starting with the dependent variable, to estimate the extensive margin, we construct three kinds of dummy variables. The first of these is NEW_EXP, which takes a value of 1 if the firm did not export to any of the regions considered in our analysis (i.e., Asia, North America, Central and South America, Africa, and Oceania) in years t - 3 to t - 1 but exported in year t. (4) The aim of using this 3-year window is to identify export starters as unambiguously as possible. Employing this definition means that export starters still include firms that have past export experience and therefore are not pure first-time exporters. However, we believe that using a 3-year window should sufficiently reduce any possible biases arising from the misidentification of new exporters. Roberts and Tybout (1997), for example, find that although a firm's export experience in the past significantly raises the odds of exporting, this experience advantage disappears within 2 years after a firm stops exporting. (5) The second, alternative dependent variable we use is NEW_EXP_REGION, which takes a value of 1 if the firm did not export to one of the regions we focus on in years t - 3 to t - 1 but did export to one of those regions in year t. The third dependent variable is defined by region. Thus, NEW_EXP_ASIA takes a value of 1 if the firm did not export to Asia in years t - 3 to t - 1 but did export to Asia in year t. Similarly, we define NEW_EXP_NA, NEW_EXP_CSA, NEW_EXP_AFR, and NEW_EXP_OCE, for the decision to export to North America, Central and South America, Africa, and Oceania, respectively.

Let us next look at our explanatory variables. The variable we are particularly interested in is the time-variant variable measuring the potential information spillovers through a main bank, BANKINFO. To construct the BANKINFO variable, we first construct the variable NUM_EXPORTER, which denotes the number of each bank's exporting client firms in each year. Note that for the NUMJEXPORTER variable, exporting firms for which a bank is not the main bank (i.e., not the top lender) are included. In this sense, we implicitly assume that all loan exposures to firms potentially contribute to the accumulation of overseas information at banks. Therefore, the NUM_EXPORTER variable measures how many firms that could serve as a source of overseas information a firm's main bank transacts with. Given that NUM EXPORTER is highly correlated with banks' size, we define BANKINFO as the ratio of NUM^EXPORTER to the total number of the bank's client firms (NUM_CLIENT) in each year. Through this metric, we aim to measure the time-variant intensity of each bank's exposure to exporting firms.

Whether a bank has branches or subsidiaries abroad and how long these overseas branches or subsidiaries have been in operation are alternative measures for banks' stock of information on overseas markets. However, in this paper, we focus on banks' transaction relationships with exporters, for the following reasons. First, Japanese banks drastically reduced the number of overseas branches at the end of the 1990s when the banking sector took drastic restructuring measures to dispose of bad debts. Instead, they increasingly engage in business tie-ups with other domestic and/or foreign banks to provide international business support services to their client firms. Therefore, we do not consider the number of banks' overseas branches to be a good proxy for the amount of information on overseas markets accumulated by banks. Second, the number of overseas branches by country or region for each bank is not readily available in the database, whereas the total number of overseas branches for each bank is available. We have to compile the data using various data sources. Nevertheless, considering alternative measures for information spillovers through banks in the future would be a worthwhile exercise.

As we have information regarding which regions each firm exports to, we can also define NUM_EXPORTER and BANKINFO by region. We assume that BANKINFO measured regardless of destination regions is a proxy for information held by banks on foreign markets in general, whereas BANKINFO measured for each destination region is a proxy for region-specific information held by banks. For each firm, we use the BANKINFO variable in order to capture the amount of information provided by the main bank. To control for the size of the main bank, we also include NUM_CLIENT in our explanatory variables. (6) To illustrate the distribution of NUM_CLIENT and BANKINFO, Table 2 presents a tabulation of the figures for these two variables as well as the BANKINFO variables for Asia and North America. Specifically, the upper part of the table shows the 20 largest sample banks in terms of NUM_CLIENT as of the end of fiscal year 1999 (FY1999), while the lower part of the table shows the average, standard deviation, maximum, and minimum of each variable measured over all sample banks in FY1999. The figures for the BANKINFO variables indicate that there is a certain degree of variation across banks. Moreover, comparing the standard deviation for each of the three BANKINFO variables in the table indicates that this variation is very similar for the three BANKINFO variables. On the other hand, in Table 1, where the basic statistics based on firm-level data are shown, the standard deviation of the region-specific BANKINFO variable is larger than the standard deviation of the BANKINFO variable for all regions. This means that using the region-specific BANKINFO variable instead of the BANKINFO variable measured for all regions increases the variation of the variable across firms.

One might argue that firms that are thinking of expanding their business overseas might try to establish a transaction relationship with a bank that is more likely to have numerous overseas information. Given that such reverse causality could generate simultaneous equation bias in our estimation, we limit the sample to firms who had the same main bank throughout year t -- 3 to year t. This allows us to focus on firm-bank pairs where the relationship is independent of the firm's decision to start exporting in year t.

As for firm-specific variables, we include variables representing firms' size, labor quality, financial constraints, own overseas activities, and productivity. For firm size, we use the (logarithm of) number of employees (LNJXUMWORKER) and for labor quality the average wage (WAGE). Regarding financial constraints, we construct a number of variables: the leverage of a firm (ratio of total liabilities to total assets, FLEV), the ratio of bank loans to total liabilities (FBDEP), the ratio of liquidity assets to liquidity liabilities (FLIQ), and the short-term loan ratio (ratio of short-term bank borrowing to total bank borrowing, STLOAN). We construct a number of variables representing firms' own overseas activities: the share of overseas establishments (FOR_BRANCH), measured as the ratio of a firm's number of overseas branches or offices (not including overseas subsidiaries or affiliates) to the firm's total number of establishments, branches, or offices, including both domestic and overseas ones; the share of overseas employees (FOR_EMP), measured as the ratio of a firm's number of workers employed in overseas branches or offices (not including overseas subsidiaries or affiliates) to the firm's total number of workers employed in all establishments, branches, or offices; the overseas investment share (FOR_INV), measured as the ratio of a firm's overseas investment (including portfolio investment) to the firm's total investment; and the overseas lending share (FOR_LOAN), measured as the ratio of a firm's lending to affiliated firms overseas to the firm's total lending to affiliated firms at home and abroad. (7)

As for firm productivity, which, as mentioned above, is widely considered to be an important determinant of the export decision, we use the firm-level TFP data provided in the East Asian Listed Companies Database (EALC) 2010. (8) The firm-level TFP in the database is calculated using the multilateral TFP index method developed by Good et al. (1997). (9)

Our firm-bank matched data cover the fiscal period from 1997 to 2008. In order to control for the potential influence of outliers, we excluded observations in the tails for each variable. (10) Table 3 shows the distribution of our sample firms by industry and year. As can be seen from the table, sample firms are concentrated in a limited number of industries (e.g., food and kindred products, chemicals, non-electrical machinery, electrical and electronic machinery, motor vehicles, transportation equipment and ordnance).

IV. ESTIMATION RESULTS

A. Decision to Enter Specific Markets

We first examine the determinants of firms' decision to participate in a new export market by estimating Equation (1). The estimation is conducted using observations for firms that did not export during the years t - 3 to t ("never" exporters) and observations for firms that did not export during the years t - 3 to t - 1 but exported in year t (first-time exporters). Thus, observations for firms that exported in at least 1 year during t - 3 to t - 1 as well as t are excluded in the estimation. The results of the random effect probit estimation (average marginal effects) are shown in Table 4. The first two columns in Table 4 show the results when we use NEW_EXP as the dependent variable and including (column 1) or excluding (column 2) TFP x BANKINFO among the explanatory variables. Columns 3 and 4 repeat the same regressions but using NEW_EXP_REGION as the dependent variable. In Columns 1-4 in Table 4, we do not distinguish between destination regions, and the BANKINFO variable is simply the ratio of the number of a firm's main bank's exporting clients--regardless of the destination region--to the total number of the bank's client firms. BANKINFO here, therefore, captures the main bank's general exposure (not specific to a destination region) to client firms with export activities. However, in the last column of Table 4, we use the region-specific BANKINFO variable corresponding to the region to which a firm starts exporting. (11)

Looking at the results shown in Table 4 and focusing on the variable of main interest, BANKINFO, we find that the coefficient is positive and significant in all cases. The estimated coefficient associated with BANKINFO is not only statistically significant but also has a sizable economic impact. For example, in the case of column 2, the coefficient (i.e., 2.0666) implies that a 1 percentage point increase in BANKINFO leads to a 2 percentage point increase in the probability of starting exports. Given that the mean of NEW_EXP in Table 1 is 0.02, the impact of BANKINFO is economically important. Similarly, in the case of column 5, where we use the BANKINFO variable measured for each destination region, the coefficient (i.e., 0.4764) implies that a 1 percentage point increase in BANKINFO leads to an almost 0.5 percentage point increase in the probability of starting exports to a new region. While the economic impact associated with a 1 percentage point increase in BANKINFO is smaller in the latter case (column 5), the standard deviation of BANKINFO measured for each destination region (0.19 in Table 1) is more than twice as large as that of BANKINFO for all regions (0.07 in Table 1). This implies that the economic impact of these two alternative BANKINFO variables is more or less comparable. Note that although we have interpreted our results as implying that the causal relationship runs from BANKINFO to firms' export dynamics, it is also conceivable that firms planning to start exporting self-select themselves into dealing with banks with higher BANKINFO. However, as explained in Section IIIB, we limit the sample to firms that had the same main bank throughout the period from year t - 3 to year t in order to mitigate the endogeneity that would arise if firms self-select into dealing with banks with numerous overseas market information.

As for the other explanatory variables, firms' own overseas activities (e.g., the overseas employee ratio) in many of the estimations have a positive effect on firms' decision to start exporting. On the other hand, the results for firm size, leverage, and liquidity vary depending on the estimation procedure.

A notable result is that the TFP level has almost no impact on the export decision. Given that the correlation between TFP and the interaction term between TFP and BANKINFO {TFP X BANKINFO) is very high, we run the same regressions without the interaction term (i.e., columns 2 and 4 in Table 4). The results remain qualitatively unchanged. While this finding is in conflict with the results of many previous studies that find a positive correlation between TFP and the export decision, it is consistent with a number of studies (e.g., Ito 2012; Todo 2011) showing that in the case of Japanese firms, TFP is not a sufficiently strong factor to explain the export decision. Of interest in this context is also the study by Wakasugi et al. (2008), which finds that productivity differences between exporting and non-exporting firms are substantially smaller in Japan than in Europe. They consequently argue that productivity differences alone do not explain whether Japanese firms engage in exports or not. A further possible reason why TFP does not appear to play a significant role is the large standard deviation of the TFP variable. As can be seen in Table 1, the average TFP level is very low, but the standard deviation is large, suggesting that the large variation in the TFP measure may be responsible for reducing the significance of the coefficient estimate for TFP. Yet another possible explanation why our result for TFP seems to conflict with previous studies is that we explicitly focus on first-time exporters, while previous studies that find a positive and significant coefficient for TFP do not necessarily strictly identify first-time exporters. Previous studies typically examine firms' export decision by looking at differences in firms' export status between the reference year and the previous year. In fact, a substantial number of firms change their export status very frequently (i.e., they keep starting and stopping exports), and identifying such export status switchers as export starters may lead to an estimation bias. The coefficient on TFP may be upward biased if firms with export experience in the recent past tend to be more productive than firms without export experience and if the former type of firms is more likely to become exporters again. In this study, we tried to strictly define first-time exporters in order to avoid the potential bias, which may have resulted in the insignificant coefficient on TFP. Moreover, the insignificant result for TFP may imply that the size or growth of demand in destination countries plays an important role in firms' export decisions, as suggested by recent studies such as Aw and Lee (2014).

Thus, there are several possible reasons why TFP is found to be insignificant in this study, and it would be hasty to conclude that productivity is not a relevant determinant in the export decision, as most previous studies on countries other than Japan argue that future export starters tend to be more productive than future non-exporters years before they enter the export market (see Wagner 2012). Note that in this study, we measure TFP using firms' revenue rather than physical output quantities, and normally one would expect revenue-based TFP to be more closely related to firms' export decision than output-based TFP. The reason is that as firm-level price deflators are usually not available, industry-level price deflators are typically used to calculate firms' revenue in real terms. Therefore, generally speaking, the revenue-based TFP measure at firm-level reflects not only firms' technological efficiency but also their markup, which tends to be higher for firms with stronger market power. Therefore, revenue-based TFP is more likely to be positively correlated with firms' export decision than quantity-based TFP, as market power is likely related to firms' profitability and their ability to cover the fixed costs to start exporting. However, despite using revenue-based TFP and not quantity-based TFP, our result for TFP is still insignificant. Thus, why TFP turns out to be insignificant in our estimation deserves further scrutiny in future studies.

Next, to examine whether the effect of region-specific information spillovers differs depending on the destination region, we split the sample by export destination region. The estimation results for the subsamples by destination region are shown in Table 5. The results suggest that BANKINFO has a significant positive effect on firms' export decision when they start exporting to Asia (column 1), but that this is not the case for other regions. These results may reflect the fact that most Japanese banks have been increasingly putting efforts into their business in Asia by expanding service networks there while restructuring services in other regions, particularly in developed regions. Furthermore, because first-time exporters to Asia tend to be smaller firms than those to other regions, the result may imply that information accumulated in main banks is more important for smaller firms, which do not have adequate capabilities to collect overseas information by themselves. Evidently, the number of employees in firms beginning to export to Asia is 1,071 on average, while that in firms exporting to other regions is 1,400. In fact, when we split the sample by firm size instead of destination region and ran the estimation for the two subsamples, we found that BANKINFO had a stronger positive effect on firms' export decision in the case of smaller firms than in larger firms. (12) Together, these results jointly support the conjecture that BANKINFO has a significant positive effect on firms' export decisions when they start exporting to Asia possibly because small first-time exporters to Asia receive a large benefit from collecting overseas information from their main banks.

B. Export Volume and Export Growth

Table 6 reports the fixed-effect panel estimation results of Equation (2). In the estimation, we only include observations of first-time exporters, and we examine whether information spillovers through main banks affect the export volume (the value of exports in logarithm) or the growth rate of exports from year t to year t + 1 after the firm started exporting. Beginning with the results in Table 6, we find that the coefficient on BANKINFO is not significant, implying that information spillovers do not have a clear effect on the volume of exports (i.e., the intensive margin). While firms' own international activities (the overseas investment ratio in column 1) tend to have a positive effect on the intensive margin, most of the other explanatory variables do not have a significant coefficient. Although it is possible that the results partly reflect the small sample size, they suggest that the export volume is mainly explained by firm fixed effects.

Next, we further split the sample by destination region and estimate the same equations as in Table 6 for each destination region. (13) The coefficient on BANKINFO is not statistically significant in all cases when the export volume is used as the dependent variable. Although we find a negative and significant coefficient on BANKINFO in the estimations for North America, Africa, and Oceania when using the export growth as the dependent variable, it should be noted that the number of observations is small, particularly in the latter two cases, for which we could not calculate F values. Therefore, we do not obtain clear and robust results for the impact of information spillovers on the intensive margin. This is in line with Koenig et al. (2010), who also do not find a significant impact of export spillovers on the intensive margin. Although our results are consistent with their results, which factors affect the intensive margin of exports is an issue that deserves further scrutiny.

C. Robustness Checks

We conducted a number of alternative estimations to check the robustness of our results. First, we estimated Equation (1) using an alternative BANKINFO variable. More specifically, we computed the weighted average of BANKINFO over all client firms for each bank using loans outstanding as of the end of each year as weights. In other words, the weighted average of BANKINFO for each bank exactly corresponds to the ratio of its total loans outstanding to exporting client firms over total loan outstanding to all its client firms. The reason for choosing loans outstanding as weights is the presumption that larger firms tend to borrow larger amounts and that firms tend to borrow larger amounts from a bank with which they have a closer relationship. Therefore, if exporting client firms tend to be larger than non-exporting client firms, the weighted average of BANKINFO will be larger than the original BANKINFO variable. Furthermore, if a bank tends to have a closer and deeper relationship with exporting client firms than non-exporting client firms, the weighted average of BANKINFO will be larger than the original BANKINFO variable. Therefore, we assume that the weighted average of BANKINFO reflects both the size of firms and the closeness of loan relations. We assume that larger exporting firms are likely to have more information on overseas markets and that banks can obtain more information from client firms with whom they have deeper loan relationships. The estimation results for NEW_EXP and NEW_EXP_REGION are shown in columns 1 and 2 in Table 7 and indicate that our main results (in Table 4) are robust to this alternative definition of the BANKINFO variable.

Second, we also estimated Equation (1) using an industry-specific BANKINFO variable. To construct this variable, we counted for each bank the number of client firms falling into particular industry categories. We then calculated the industry-specific BANKINFO variable for each bank by taking the ratio of the number of exporting clients in a particular industry to the total number of clients in that industry. Note that for this exercise, we focus on four industries, for which there are a sufficient number of firms for each industry. The industries are (1) food and kindred products, (2) chemicals, (3) electrical and electronic machinery, and (4) motor vehicles, transportation equipment, and ordnance. The reason for restricting the analysis to these four industries is that the number of observations for other industries is too small, so that the standard deviation of the BANKINFO variable becomes very large, which reduces the significance level of the estimation results. The estimation results for NEW_EXP and NEW_EXP_REGION using the BANKINFO variable thus defined are presented in columns 3 and 4 in Table 7 and confirm that our main results are robust to this alternative definition of the BANKINFO variable.

Third, we also estimated Equation (1) using a logit estimator, for which the standard errors are corrected for clustering. Taking into account that observations within the same firm are not independent, standard errors are corrected for clustering across firms. Alternatively, standard errors are corrected for clustering across main banks, considering the possibility that observations of firms that have a transaction relationship with the same bank are not independent. Furthermore, we also cluster the standard errors within firm-bank pairs to account for possible correlations between the export behavior of a firm having a particular main bank over time. In all the cases, the results are consistent with the results in Table 4, and BANKINFO has a significant positive effect on firms' export decision. (14)

In addition, bank characteristics may affect firms' export decision. For example, the Japan Bank for International Corporation (JBIC) is a government financial institution, which was originally established to promote cross-border trade and foreign investment. Therefore, JBIC may be particularly active in helping firms to start exporting. On the other hand, major commercial banks may differ from regional banks or local banks in terms of their scope of business and hence in the characteristics of information accumulated by them. In order to control for differences in bank characteristics, we include a JBIC dummy and a dummy for major commercial banks in the export decision estimation. However, neither dummy variable has a significant coefficient nor does including these dummy variables changes the significance of the BANKINFO variable.

While we lag all the explanatory variables by 1 year in our estimations, we also estimate the equations in Table 4 taking 2- or 3-year lags for robustness checks. Taking a longer lag means taking a firm's dynamic decisions into account, as well as being able to address the potential endogeneity between the export decision and firm-bank matching. We obtain similar results to those in Table 4 in the case of the 2-year lag. On the other hand, in the case of the 3-year lag, the estimated marginal effect of BANKINFO becomes insignificant when using NEW_EXP as the dependent variable. These results may suggest that firms begin to prepare 2 years in advance for starting to export. As firms' export decision may be often made in response to short-term demand fluctuations, it may be too early to prepare for exporting 3 years in advance.

Finally, there may be several alternative ways to measure the amount of information on export markets available to a firm. While our main variable, BANKINFO, measures the intensity of banks' exposure to exporting firms, the absolute number of a bank's exporting client firms, NUM_EXPORTER may be a better way to measure the amount of information on export markets. Given this, we estimate the same model as Equation (2) in Table 4 including NUM_EXPORTER instead of BANKINFO. We obtain positive and significant coefficient on NUM_EXPORTER.

We also conduct two additional estimations for the robustness check using subsamples of our sample firms. First, due to the manner of calculating BANKINFO, we might wrongly measure each bank's exposure to export market information when the number of client firms for a bank is very small. In fact, some of the regional banks have a limited number of listed companies as their clients. In such a case, BANKINFO might not work as a good proxy for the information related to export markets. Thus, we limit our sample firms to firms whose main bank has more than 30 (first and third columns) or 50 (second and fourth columns) client firms (NUM_CLIENT).

The results are consistent with those presented in Table 4, which confirms the robustness of our results. Second, BANKINFO could largely vary due to mergers and acquisitions (M&As) among banks. In fact, a certain amount of variation in BANKINFO in the 2000s is driven by M&As among banks. In order to exclude the effect of banks' M&As, we repeat the same estimations as those in Table 4 excluding the observations of firms whose main banks experienced M&As. The results show consistent results with our baseline estimations.

V. CONCLUDING REMARKS

In this paper, we examined whether information spillovers through main banks affect client firms' export behavior (i.e., the extensive and intensive margins). We find that information spillovers through main banks have a positive effect on client firms' decision to start exporting. This implies that information on destination markets provided by main banks substantially reduces the fixed entry cost of exporting and encourages firms to become exporters. However, we did not find evidence that information spillovers through main banks have an effect on the export volume or on the growth rate of exports. This is more or less consistent with the findings obtained by Koenig et al. (2010).

A key contribution of this paper is that it proposes an additional channel of information spillovers ignored in previous studies. While existing studies, such as Koenig et al. (2010), concentrate on information spillovers from other exporting firms in the same region and/or industry, this study focuses on the importance of information provided directly by main lender banks through transaction relationships. If we look at our results in terms of the argument put forward by Chaney (2008) that a change in fixed costs affects only the extensive margin, whereas a change in variable costs affects both the intensive and extensive margins, they suggest that information provided by banks contributes to a reduction in the fixed costs but not in the variable costs associated with exporting. Conversely, Paravisini etal. (2015) suggest that credit frictions, by affecting the cost of working capital, affect the variable costs of exporting and hence the volume of exports. This result suggests that banks may play an important role in affecting the intensive margin as suppliers of funds. Thus, banks' role as providers of information and as suppliers of funds may affect fixed and variable costs and hence the extensive and intensive margin differently. Untangling these two roles of banks and their impact on firms' export behavior is a topic we aim to further address in future research.

This paper also provides an important policy implication. As mentioned in Section I, our knowledge regarding what factors are important for firms to become an exporter remains very limited, even though export promotion has been an important policy issue in many countries. With regard to Japan, studies such as those by Wakasugi et al. (2008) and Ito (2012) argue that many firms still do not export even though their performance is good or they actively invest in research and development. Promoting exports by these firms is an urgent policy issue for Japan, which has been facing population decline and sluggish domestic demand for a prolonged period. This paper showed the importance of banks' role as an information provider for potential exporters, implying that the government should proactively involve banks in its export promotion policies. Regional banks--seeing their client firms face declining domestic demand and therefore worried that their own business may shrink--may also be interested in providing more support services for firms trying to expand their business abroad. Helping such banks to build international service networks and building on the banks' support services may allow the government to implement its export promotion policies more effectively. Moreover, as banks have accumulated numerous information on their client firms' business, they may have useful knowledge on what type of firms should receive support from the government and on what type of support is most effective. Of course, government and non-profit organizations already provide various support services for firms' international business and for trading companies. Information provided by such organizations or trading companies is complementary to information collected by banks through lending relationships, and it is important for the government to effectively utilize these various information sources for export promotion policies. According to the banker we interviewed, the advantage that banks have is that they possess detailed and wide-ranging information on individual firms' management, financial health, and business activities.

To conclude, we highlight several issues for future research. First, our long panel data set allows us to conduct a survival analysis-type study on the status of exporting firms. This, in turn, allows us to explicitly examine the shape of hazard function of firms exiting from export markets. This is another important dimension discussed in the theoretical international trade literature (e.g., Schroder and Sprensen 2012). Although a fair number of empirical studies have analyzed the determinants of international trade relationships duration using survival analysis, most of these studies have analyzed the duration at an aggregate or product-country level (e.g., Besedes and Blyde 2010; Besedes and Prusa 2006a, 2006b; Nitsch 2009). Very few studies have examined the determinants of survival in export markets through hazard estimation using firm-level data, with the exception of Esteve-Perez et al. (2013). Second, although the expansion of export destinations, particularly in the case of larger listed firms, often involves the establishment of new subsidiaries or affiliates abroad, this paper, partly because of data constraints, focused only on exporting and did not explicitly deal with foreign direct investment in a new location. As banks provide a wide range of support services for firms that try to open a foreign affiliate, investigating banks' role in firms' foreign direct investment decision is another promising research topic. Lastly, our results imply that information spillovers through main banks may be more important for smaller firms, which are more likely to choose Asia as their first export destination. Therefore, further investigation focusing on smaller firms would be a worthwhile exercise, if data for small firms were available. We believe that all of these extensions would provide further evidence for a better understanding of firms' overseas activities and the role of banks.

ABBREVIATIONS

BSBSA: Basic Survey on Business Structure and Activities

EALC: East Asian Listed Companies Database

FY: Fiscal Year

JBIC: Japan Bank for International Corporation

M&As: Mergers and Acquisitions

METI: Ministry of Economy, Trade and Industry

NEEDS: Nikkei Economic Electronic Database System

TFP: Total Factor Productivity

doi: 10.1111/ecin. 12211

Online Early publication March 13, 2015

REFERENCES

Aitken, H., G. H. Hanson, and A. E. Harrison. "Spillovers, Foreign Investment, and Export Behavior." Journal of International Economics, 43(1-2), 1997, 103-32.

Amiti, M., and D. Weinstein. "Exports and Financial Shocks." Quarterly Journal of Economics, 126(4), 2011, 1841-77.

Aw, B. Y., and Y. Lee. "A Model of Demand, Productivity and Foreign Location Decision Among Taiwanese Firms." Journal of International Economics, 92(2), 2014. 304-16.

Aw, B. Y, S. Chung, and M. L Roberts. "Productivity and Turnover in the Export Market: Micro-level Evidence from Taiwan (China) and the Republic of Korea." World Bank Economic Review, 14(1), 2000, 65-90.

Barrios, S., H. Gorg, and E. Strobl. "Explaining Firms' Export Behaviour: R&D, Spillovers and the Destination Market." Oxford Bulletin of Economics and Statistics, 65(4), 2003,475-96.

Bernard, A. B., and J. B. Jensen. "Why Do Firms Export." Review of Economics and Statistics, 86(2), 2004, 561-9.

Bernard, A. B., J. Eaton, J. B. Jensen, and S. Kortum. "Plants and Productivity in International Trade." American Economic Review, 93,2003, 1268-90.

Besedes, T., and J. Blyde. "What Drives Export Survival? An Analysis of Export Duration in Latin America." Mimeo, Inter-American Development Bank, Washington, DC, 2010.

Besedes, T., and T. J. Prusa. "Product Differentiation and Duration of U.S. Import Trade." Journal of International Economics, 70, 2006a, 339-58.

--. "Ins, Outs, and the Duration of Trade." Canadian Journal of Economics, 39(1), 2006b, 266-95.

Chaney, T. "Distorted Gravity: The Intensive and Extensive Margins of International Trade." American Economic Review, 98(4), 2008, 1707-21.

Creusen, H., and A. Lejour. "Uncertainty and the Export Decisions of Dutch Firms." CPB Discussion Paper 183, CPB Netherlands Bureau for Economic Policy Analysis, 2011.

Degryse, H., M. Kim, and S. Ongena. Microeconometrics of Banking: Methods, Applications, and Results. Oxford: Oxford University Press, 2009.

Esteve-Perez, S., F. Requena-Silvente, and V. J. Pallardo-Lopez. "The Duration of Firm-Destination Export Relationships: Evidence from Spain 1997-2006." Economic Inquiry, 51(1), 2013, 159-80.

Feenstra, R. C., Z. Li, and M. Yu. "Exports and Credit Constraints under Incomplete Information: Theory and Evidence from China." Review of Economics and Statistics, 96(4), 2014, 729-44.

Fukao, K., T. Inui, K. Ito, Y. G. Kim, and T. Yuan. "An International Comparison of the TFP Levels and the Productivity Convergence of Japanese, Korean, Taiwanese, and Chinese Listed Firms." Journal of Chinese and Economic and Business Studies, 9(2), 2011, 127-50.

Good, D., M. Nadiri, and R. Sickles. "Index Number and Factor Demand Approaches to the Estimation of Productivity," in Handbook of Applied Econometrics: Microeconomics, edited by H. Pesaran and P. Schmidt. Oxford: Blackwell, 1997, 14-80.

Ito, K. "Sources of Leaming-by-Exporting Effects: Does Exporting Promote Innovation?" Working Papers DP2012-06, Economic Research Institute for ASEAN and East Asia (ERIA), 2012.

Koenig, P, F. Mayneris, and S. Poncet. "Local Export Spillovers in France." European Economic Review, 54, 2010. 622-41.

Krautheim, S. "Heterogeneous Firms, Exporter Networks and the Effect of Distance on International Trade." Journal of International Economics, 87, 2012, 27-35.

Manova, K., S. Wei, and Z. Zhang. Forthcoming. "Firm Exports and Multinational Activity under Credit Constraints." Review of Economics and Statistics.

Mayer, T., and G. Ottaviano. "The Happy Few: The Internationalisation of European Firms." Intereconomics, 43, 2008, 135-48.

Minetti, R., and S. C. Zhu. "Credit Constraints and Firms Export: Microeconomic Evidence from Italy." Journal of International Economics, 83, 2011, 109-25.

Nitsch, V. "Die Another Day: Duration in German Import Trade." Review of World Economics, 145, 2009, 133-54.

Paravisini, D., V. Rappoport, P. Schnabl, and D. Wolfenzon. "Dissecting the Effect of Credit Supply on Trade: Evidence from Matched Credit-Export Data." Review of Economic Studies, 82(1), 2015, 333-59.

Patrick, H. "The Relevance of Japanese Finance and Its Main Bank System," in The Japanese Main Bank System: Its Relevance for Developing and Transforming Economies, edited by M. Aoki and H. Patrick. Oxford: Oxford University Press, 1994, 359-60.

Rauch, J. E., and J. Watson. "Starting Small in an Unfamiliar Environment." International Journal of Industrial Organization, 21(7), 2003, 1021-42.

Roberts, M., and J. Tybout. "The Decision to Export in Colombia: An Empirical Model of Entry with Sunk Costs." American Economic Review, 87(4), 1997, 545-64.

Rodrik, D. "The Dynamics of Political Support for Reform in Economies in Transition." Journal of the Japanese and International Economies, 9(4), 1995, 403-25.

Schroder, P. J. H., and A. Sorensen. "Firm Exit, Technological Progress and Trade." European Economic Review, 56, 2012. 579-91.

Todo, Y. "Quantitative Evaluation of Determinants of Export and FDI: Firm-Level Evidence from Japan." The World Economy, 34(3), 2011, 355-81.

Wagner, J. "International Trade and Firm Performance: A Survey of Empirical Studies Since 2006." Review of World Economics, 148, 2012, 235-67.

Wakasugi, R., Y. Todo, H. Sato, S. Nishioka, T. Matsuura, B. Ito, and A. Tanaka. "The Internationalization of Japanese Firms: New Findings Based on Firm-Level Data." RIETI Discussion Paper 08-E-036, Research Institute of Economy, Trade and Industry, 2008.

Inui: Professor, Preparatory Office for the Faculty of International Social Studies, Gakushuin University, Toshima-ku, Tokyo 171-8588, Japan. Faculty Fellow, Research Institute of Economy, Trade and Industry, Chiyoda-ku, Tokyo 100-8901, Japan. Phone 81 3-5992-1432, Fax 81 3-5992-1432, E-mail tomohiko.inui@gakushuin.ac.jp

Ito: Professor, School of Economics, Senshu University, Kawasaki, Kanagawa 214-8580, Japan. Phone 81-44-900-7818, Fax 81-44-911-0467, E-mail keikoi@isc.senshu-u.ac.jp

Miyakawa: Associate Professor, College of Economics, Nihon University, Chiyoda-ku, Tokyo 101-8360, Japan. Phone 81-3-3219-3468, Fax 81-3-3219-3468, E-mail miyakawa.daisuke@nihon-u.ac.jp

(1.) A more recent study making a similar argument that repeated loan transactions lead to the accumulation of soft information on client firms, by Degryse et al. (2009).

(2.) Of course, there are several other sources from which firms obtain information on export markets. Economic diplomacy and chambers of commerce in destination countries (Creusen and Lejour 2011) are another source of information on foreign markets, although we do not address the role of economic diplomacy here due to data constraints. Yet another potentially important conduit of information on export markets is trading companies and wholesalers. Unfortunately, we cannot identify transaction relationships between exporter firms and trading companies.

(3.) The survey asks for the amount as well as the destination or origin of exports and imports broken down into seven regions (Asia, Middle East, Europe, North America, Latin America, Africa, and Oceania). Unfortunately, more detailed information on the destination of exports and origin of imports is not available.

(4.) The BSBSA also specifies other destination regions such as the Middle East and Europe. We ignore these regions due to the small number of export starters to those regions.

(5.) We are grateful to an anonymous referee for pointing this out to us.

(6.) While NUM_CLIENT represents the size of a bank, size is highly correlated with banks' financial health and performance such as efficiency or profitability. Therefore, by including NUM_CLIENT in the explanatory variables, we intend to control not only for bank size but also for various other bank characteristics.

(7.) Note that, in our data set, a limited number of firms belonging to specific industries (e.g., food, paper, and chemicals) own overseas branches but have not started exports yet. On the other hand, a significant number of firms have invested in, or lent money to local companies overseas. The aim of including firms' own overseas activities is to control for these specific characteristics of firms.

(8.) The EALC is jointly compiled by the Japan Center for Economic Research, the Center for Economic Institutions (Hitotsubashi University), the Center for China and Asian

Studies (Nihon University), and the Center for National Competitiveness (Seoul National University).

(9.) For more details on the calculation of firm-level TFP used here, see Fukao et al. (2011).

(10.) We drop firms for which the absolute level of any of the explanatory variables falls into the 1st or the 99th percentile.

(11.) In the case where firms start exporting to more than one region at a time, we randomly assign the region-specific BANKINFO. An alternative way would be to use the average of BANKINFO among those regions.

(12.) The results of the subsample estimations are available from he authors upon request.

(13.) The estimation results for the subsamples are suppressed in order to conserve space and are available from the authors upon request.

(14.) The results of the rest of the estimations in this section are available from the authors upon request. TABLE 1 Summary Statistics Variable Observations Average NEW EXP (dummy for starting export) 3,220 0.02 NEW EXP REGION (dummy for expanding 3,220 0.15 export region) NEW EXP ASIA (dummy for starting 3,220 0.03 Asian export) NEW EXP NA (dummy for starting 3,220 0.03 Northern America export) NEW EXP CSA (dummy for starting 3,220 0.07 Southern & Central America export) NEW EXP OCE (dummy for starting 3,220 0.04 Oceania export) LN NUMWORKER (log of the number of 2,914 7.02 workers) FLEV (total liability/total asset) 3,205 0.52 FBDEP (bank loan/total liability) 3,209 0.31 FLIQ (liquidity asset/liquidity 3,215 1.56 liability) STLOAN (short-term bank loan/bank 2,948 0.53 loan) WAGE (labor quality: average wage) 2,903 6.49 FOR BRANCH (ratio of foreign cities) 3,206 0.05 FOR EMP (ratio of workers abroad) 3,206 0.002 FOR INV (ratio of investments abroad) 3,201 0.25 FOR LOAN (ratio of lending abroad) 3,220 0.11 TFP (productivity) 2,780 0.02 NUM EXPORTER (number of clients 3,190 182.90 exporting) NUM CLIENT (number of clients) 3,190 353.06 BANKINFO (NUM EXPRTER/NUM CLIENT) 3,190 0.52 BANKINFO (measured for the 3,176 0.08 destination region) Variable Standard Minimum Maximum Deviation NEW EXP (dummy for starting export) 0.15 0 1 NEW EXP REGION (dummy for expanding 0.36 0 1 export region) NEW EXP ASIA (dummy for starting 0.17 0 1 Asian export) NEW EXP NA (dummy for starting 0.17 0 1 Northern America export) NEW EXP CSA (dummy for starting 0.25 0 1 Southern & Central America export) NEW EXP OCE (dummy for starting 0.2 0 1 Oceania export) LN NUMWORKER (log of the number of 1.11 4.03 10.59 workers) FLEV (total liability/total asset) 0.18 0.05 0.96 FBDEP (bank loan/total liability) 0.21 0 0.89 FLIQ (liquidity asset/liquidity 0.85 0.26 8.46 liability) STLOAN (short-term bank loan/bank 0.32 0 1 loan) WAGE (labor quality: average wage) 1.78 0.46 12.72 FOR BRANCH (ratio of foreign cities) 0.11 0 0.68 FOR EMP (ratio of workers abroad) 0.01 0 0.07 FOR INV (ratio of investments abroad) 0.44 0 3.36 FOR LOAN (ratio of lending abroad) 0.26 0 1 TFP (productivity) 0.11 -0.97 0.59 NUM EXPORTER (number of clients 92.41 1 371 exporting) NUM CLIENT (number of clients) 183.63 8 759 BANKINFO (NUM EXPRTER/NUM CLIENT) 0.07 0.08 0.78 BANKINFO (measured for the 0.19 0.00 3.67 destination region) TABLE 2 Distribution of NUM_CLIENT and BANKINFO in 1999 Fiscal Year 1999 Bank ID NUM_CLIENT BANKINFO BANKINFO BANKINFO Asia North America 1 518 0.62 0.59 0.52 2 390 0.58 0.55 0.49 3 376 0.57 0.55 0.49 4 353 0.58 0.56 0.48 5 346 0.55 0.52 0.46 6 338 0.55 0.51 0.46 7 337 0.55 0.52 0.46 8 267 0.57 0.54 0.48 9 256 0.51 0.50 0.43 10 252 0.57 0.54 0.46 11 233 0.55 0.53 0.45 12 223 0.51 0.48 0.43 13 217 0.62 0.59 0.53 14 190 0.51 0.49 0.41 15 179 0.56 0.53 0.49 16 176 0.56 0.55 0.44 17 162 0.57 0.56 0.48 18 140 0.68 0.64 0.61 19 135 0.54 0.49 0.42 20 108 0.60 0.58 0.54 All banks in our sample Number of 49 49 49 49 observations Average 169 0.52 0.50 0.43 Standard deviation 136 0.10 0.09 0.10 Maximum 518 0.68 0.67 0.61 Minimum 9 0.17 0.17 0.11 TABLE 3 Distribution of Sample Firms by Industry and Year 2000 2001 2002 2003 2004 Food and kindred products 43 41 40 32 32 Textile mill products, apparel 18 23 20 17 13 Lumber and wood products, 2 2 0 0 0 furniture and fixtures Paper and allied products 9 9 9 9 10 Printing, publishing, and 7 5 6 5 4 allied products Chemicals 31 30 31 25 36 Petroleum and coal products 2 1 3 2 0 Rubber and miscellaneous 6 4 7 5 5 plastics Stone, clay, and glass 13 13 16 15 17 products Metal 10 12 14 9 11 Non-metallic mining 11 8 7 6 5 Fabricated metal 15 15 14 9 11 Non-electrical machinery 18 15 13 12 19 Electrical and electronic 52 45 51 39 49 machinery Motor vehicles, transportation 28 36 31 28 36 equipment, and ordnance Instruments 7 8 5 3 4 Miscellaneous manufacturing 19 18 17 19 16 Total 291 285 284 235 268 2005 2006 2007 2008 Total Food and kindred products 34 44 44 52 362 Textile mill products, apparel 18 22 24 22 177 Lumber and wood products, 0 1 1 1 7 furniture and fixtures Paper and allied products 8 13 13 12 92 Printing, publishing, and 5 9 9 10 60 allied products Chemicals 41 49 47 51 341 Petroleum and coal products 0 2 1 1 12 Rubber and miscellaneous 5 4 7 10 53 plastics Stone, clay, and glass 16 18 21 21 150 products Metal 9 21 21 23 130 Non-metallic mining 6 12 12 15 82 Fabricated metal 10 20 19 19 132 Non-electrical machinery 24 26 35 32 194 Electrical and electronic 62 65 75 77 515 machinery Motor vehicles, transportation 43 44 46 46 338 equipment, and ordnance Instruments 3 3 7 8 48 Miscellaneous manufacturing 17 20 21 22 169 Total 301 373 403 422 2,862 TABLE 4 Random-Effect Panel Probit Estimation Results for Extensive Margin (3) (1) (2) NEW_EXP_ NEW_EXP NEW_EXP REGION Extensive Margin dy/dx dy/dx dy/dx LN_NUMWORKER 0.0594 0.0612 0.0849 ** (0.0675) (0.0672) (0.0374) FLEV 0.3496 0.3010 0.3927 (0.6523) (0.6510) (0.3297) FBDEP 0.8656 * 0.7559 * 0.0266 (0.4495) (0.4435) (0.2334) FLIQ 0.3966 *** 0.3785 *** -0.0478 (0.1466) (0.1473) (0.0734) STLOAN 0.2612 0.3073 0.0411 (0.2383) (0.2377) (0.1133) WAGE -0.0330 -0.0349 0.0068 (0.0416) (0.0416) (0.0218) FOR_BRANCH 0.5277 0.5627 -0.6871 (1.1716) (1.1886) (0.4553) FOR_EMP 24.5621 21.5684 16.4349 ** (15.5615) (16.1527) (6.5394) FOR_INV 0.2521 0.2648 -0.0238 (0.2245) (0.2179) (0.0889) FOR_LOAN -0.5484 * -0.5297 0.0226 (0.3287) (0.3291) (0.1218) TFP -10.8578 ** -0.4327 -1.2803 (5.3428) (0.8626) (3.3607) BANKINFO (a) 2.7098 *** 2.0666 ** 1.5565 ** (0.9117) (0.8510) (0.6591) TFP X BANKINFO (a) 19.4209 ** 2.8644 (9.7683) (6.3235) NUM CLIENT 0.0008 * 0.0007 0.0001 (0.0005) (0.0005) (0.0002) Number of observations 1,178 1,178 2,589 Number of groups 304 304 562 Observations per group: 1 1 1 minimum Average 3.9 3.9 4.6 Maximum 10 10 9 Wald [chi square] 56.62 54.74 232.58 Probability > [chi square] .0265 .0303 .0000 Log likelihood -313.15 -315.27 -942.19 Likelihood ratio test of 5.23 5.53 1.83 [[rho].sub.0] = 0 Probability [greater than or .011 .009 .088 equal to] [bar.[chi square]] Year dummies Yes Yes Yes Industry dummies Yes Yes Yes (4) (5) NEW_EXP_ NEW_EXP_ REGION REGION Extensive Margin dy/dx dy/dx LN NUMWORKER 0.0853 ** 0.0890 ** (0.0374) (0.0369) FLEV 0.3858 0.3923 (0.3290) (0.3237) FBDEP 0.0231 0.0250 (0.2332) (0.2290) FLIQ -0.0484 -0.0456 (0.0733) (0.0725) STLOAN 0.0447 0.0383 (0.1129) (0.1117) WAGE 0.0066 0.0111 (0.0218) (0.0216) FOR_BRANCH -0.6884 -0.6460 (0.4552) (0.4491) FOR_EMP 16.4852 ** 15.5256 ** (6.5388) (6.4744) FOR_INV -0.0251 -0.0140 (0.0888) (0.0869) FOR_LOAN 0.0215 0.0315 (0.1217) (0.1203) TFP 0.2251 -0.0084 (0.4695) (0.4941) BANKINFO (a) 1.5628 ** 0.4764 ** (0.6597) (0.2028) TFP X BANKINFO (a) 3.3046 (2.2393) NUM CLIENT 0.0001 0.0001 (0.0002) (0.0002) Number of observations 2,589 2,570 Number of groups 562 561 Observations per group: 1 4 minimum Average 4.6 4.6 Maximum 9 9 Wald [chi square] 232.48 239.03 Probability > [chi square] .0000 .0000 Log likelihood -942.29 -933.58 Likelihood ratio test of 1.8 0.61 [[rho].sub.0] = 0 Probability [greater than or .09 .217 equal to] [bar.[chi square]] Year dummies Yes Yes Industry dummies Yes No Note: Standard errors are in parentheses. (a) The BANKINFO variable for columns 1-4 is measured regardless of the destination region, while the BANKINFO variable in column 5 is measured for each destination region. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively. TABLE 5 Random-Effect Panel Probit Estimation Results for Extensive Margin by Destination Region (1) (2) (3) NEW_EXP_ NEW_EXP_ NEW_EXP_ Extensive Margin ASIA dy/dx NA dy/dx CSA dy/dx LN_NUMWORKER 0.0581 0.4464 *** 0.1009 * (0.0823) (0.1621) (0.0545) FLEV 0.7978 1.2477 1.0709 ** (0.7941) (1.1094) (0.5092) FBDEP 0.4545 1.8494 ** -0.2969 (0.5209) (0.9449) (0.3508) FLIQ 0.3822 ** 0.3905 0.1073 (0.1888) (0.2378) (0.1112) STLOAN 0.3607 0.4460 0.0302 (0.2663) (0.3702) (0.1705) WAGE -0.0798 -0.0997 0.0268 (0.0511) (0.0658) (0.0327) FOR_BRANCH 0.0332 -2.4972 -0.3675 (1.5995) (2.1615) (0.7125) FOR_EMP 42.2748 ** 77.6527 ** 17.8788 * (21.2118) (31.7772) (9.4868) FOR_INV -0.5063 0.5267 -0.0772 (0.3865) (0.3795) (0.1476) FOR_LOAN 0.0485 -0.7049 0.3178 * (0.3312) (0.5758) (0.1718) TFP -0.5318 -5.0289 -1.1761 (7.3806) (5.7492) (2.3737) BANKINFO (a) 2.8382 ** 0.6886 1.4655 (1.4160) (1.5599) (1.1103) TFP x BANKINFO (a) 1.7274 14.4149 8.8588 (13.7284) (13.6617) (9.3470) NUM_CLIENT 0.0008 0.0004 0.0001 (0.0006) (0.0007) (0.0004) Number of observations 815 1,143 1,910 Number of groups 213 275 483 Observations per group: 1 1 1 minimum Average 3.8 4.2 4 Maximum 9 9 9 Wald [chi square] 41.33 22.84 164.84 Probability > [chi square] .249 .9672 .0000 Log likelihood -157.39564 -197.99 -453.62 Likelihood ratio test of 0 7.25 1.46 [[rho].sub.0] = 0 Probability [greater than or 1 .004 .113 equal to] [[bar.[chi square] Year dummies Yes Yes Yes Industry dummies Yes Yes Yes (4) (5) NEW_EXP_ NEW_EXP_O Extensive Margin AFR dy/dx CE dy/dx LN NUMWORKER 0.1161 0.1499 ** (0.0725) (0.0763) FLEV 1.5337 ** 0.2956 (0.6847) (0.6552) FBDEP -0.3693 -0.3018 (0.4504) (0.4636) FLIQ 0.1309 -0.2702 * (0.1502) (0.1526) STLOAN -0.0180 -0.0437 (0.2283) (0.2199) WAGE 0.1001 ** 0.0641 (0.0420) (0.0410) FOR_BRANCH 0.8275 0.2359 (0.7787) (0.8553) FOR_EMP -1.5137 6.7518 (11.5536) (11.3327) FOR_INV 0.1835 0.2356 (0.1654) (0.1624) FOR_LOAN -0.0075 0.0036 (0.2206) (0.2387) TFP 0.6944 1.2884 (1.0451) (2.3621) BANKINFO (a) 0.0570 1.0355 (0.2779) (1.0289) TFP x BANKINFO (a) -1.0604 -6.3045 (2.2947) (9.1479) NUM_CLIENT 0.0002 0.0002 (0.0005) (0.0004) Number of observations 1,861 1,969 Number of groups 458 454 Observations per group: 1 1 minimum Average 4.1 4.3 Maximum 9 9 Wald [chi square] 97 40.6 Probability > [chi square] .0000 .3147 Log likelihood -352.62 -346.42 Likelihood ratio test of 6.91 4.39 [[rho].sub.0] = 0 Probability [greater than or .004 .018 equal to] [[bar.[chi square] Year dummies Yes Yes Industry dummies Yes Yes Note: Standard errors are in parentheses. (a) The BANK1NFO variable is measured for each destination region. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively. TABLE 6 Fixed-Effect Panel Estimation Results for Intensive Margin (1) (2) LN_ [DELTA]LN_ EXPORT EXPORT Intensive Margin Coefficient Coefficient All regions LN NUMWORKER 0.1596 -0.4597 *** (0.2083) (0.1744) FLEV -0.3610 -1.0596 * (0.6445) (0.5894) FBDEP -0.2657 0.0972 (0.3539) (0.3608) FLIQ -0.0557 0.0253 (0.1307) (0.1336) STLOAN -0.0966 0.0247 (0.1408) (0.1402) WAGE 0.0192 -0.0129 (0.0271) (0.0275) FOR_BRANCH 0.7586 0.1181 (0.4661) (0.4290) FOR_EMP 7.4139 -0.1965 (5.5907) (5.6848) FOR_INV 0.4138 ** 0.0531 (0.1917) (0.1541) FOR_LOAN 0.0486 0.0039 (0.0874) (0.0798) TFP 0.1745 -3.1451 (2.2030) (2.0943) BANKINFO -0.3234 -0.5169 (0.4680) (0.6068) TFP x BANKINFO -0.5957 6.9974 (4.3403) (4.2042) NUM_CLIENT 0.0001 0.0000 (0.0003) (0.0003) _cons 7.2424 *** 4.0855 *** (1.7397) (1.4323) Number of 1.656 1,328 observations Number of groups 426 389 Observations per 1 1 group: minimum Average 3.9 3.4 Maximum 9 9 F 4.7 1.91 Probability > F 0 .011 [R.sup.2]: within .0872 .03 [R.sup.2]: between .3209 .0169 [R.sup.2]: overall .247 .0028 corr(u_i, Xb) 0.3668 -0.7657 Year dummies Yes Yes Industry dummies No No Note: Standard errors clustered within a firm are in parentheses. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively. TABLE 7 Random-Effect Panel Probit Estimation Results for Extensive Margin with Alternative BANKINFO Variables Loan-Weighted (1) (2) NEW_EXP NEW_EXP_ Extensive Margin dy/dx REGION dy/dx LN_NUMWORKER 0.0641 0.0772 ** (0.0702) (0.0375) FLEV 0.5310 0.4399 (0.6815) (0.3284) FBDEP 0.8424 * 0.0112 (0.4674) (0.2313) FLIQ 0.4286 *** -0.0380 (0.1562) (0.0733) STLOAN 0.2366 0.0232 (0.2478) (0.1131) WAGE -0.0250 0.0094 (0.0431) (0.0217) FOR_BRANCH 0.5196 -0.6818 (1.2277) (0.4546) FOR_EMP 23.3913 16.3116 ** (16.0627) (6.5164) FOR_INV 0.2179 -0.0242 (0.2348) (0.0884) FOR_LOAN -0.5206 0.0307 (0.3374) (0.1216) TFP -6.1381 ** -1.4182 (2.8393) (1.3869) BANKINFO (a) 1.8253 * 0.9578 ** (0.9610) (0.4498) TFP X BANKINFO (a) 19.1933 ** 5.7667 (9.0637) (4.4617) NUM_CLIENT 0.0006 -0.0001 (0.0005) (0.0002) Number of observations 1,177 2,588 Number of groups 304 562 Observations per group: minimum 1 1 Average 3.9 4.6 Maximum 10 9 Wald [chi square] 51.87 235.65 Probability > [chi square] .0662 .0000 Log likelihood -313.94 -941.66 Likelihood ratio test of 7.28 1.50 [[rho].sub.0] = 0 Probability [greater than or equal to] .003 .110 [bar.[chi square]] Year dummies Yes Yes Industry dummies Yes Yes Industry-Specific (3) (4) NEW_EXP NEW_EXP_ Extensive Margin dy/dx REGION dy/dx LN_NUMWORKER 0.3479 ** 0.0426 (0.1669) (0.0482) FLEV 1.3753 0.5728 (1.4265) (0.4250) FBDEP 0.0006 -0.4836 (1.0120) (0.3071) FLIQ 0.4475 -0.0820 (0.3850) (0.0885) STLOAN 0.1700 0.0687 (0.4970) (0.1382) WAGE -0.1712 ** 0.0160 (0.0873) (0.0286) FOR_BRANCH -10.9655 * -1.1284 * (6.3630) (0.6845) FOR_EMP 147.7707 ** 23.8920 *** (67.8846) (8.9274) FOR_INV -0.3148 0.0181 (0.5861) (0.0989) FOR_LOAN -0.4421 0.0338 (0.6381) (0.1517) TFP 6.9263 1.4878 (6.1546) (2.0143) BANKINFO (a) 4.1028 ** 0.7766 * (1.6148) (0.4108) TFP X BANKINFO (a) -7.2430 -1.6595 (8.9337) (2.7527) NUM_CLIENT 0.0022 * 0.0003 (0.0011) (0.0003) Number of observations 539 1,371 Number of groups 146 293 Observations per group: minimum 1 1 Average 3.7 4.7 Maximum 10 9 Wald [chi square] 19.17 126.72 Probability > [chi square] .8293 .0000 Log likelihood -127.84 -512.34 Likelihood ratio test of 9.67 0.15 [[rho].sub.0] = 0 Probability [greater than or equal to] .001 .351 [bar.[chi square]] Year dummies Yes Yes Industry dummies Yes Yes Note: Standard errors are in parentheses. (a) The BANKINFO variable in columns 1 and 2 is the weighted average of BANKINFO measured regardless of the destination region using loans outstanding as weights. The BANKINFO variable in columns 3 and 4 is the industry-specific BANKINFO value (not weighted). ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively.